Communications Research Laboratory, Japan
Abstract:Recently, Ainsworth et al. showed that using weight matching (WM) to minimize the $L_2$ distance in a permutation search of model parameters effectively identifies permutations that satisfy linear mode connectivity (LMC), in which the loss along a linear path between two independently trained models with different seeds remains nearly constant. This paper provides a theoretical analysis of LMC using WM, which is crucial for understanding stochastic gradient descent's effectiveness and its application in areas like model merging. We first experimentally and theoretically show that permutations found by WM do not significantly reduce the $L_2$ distance between two models and the occurrence of LMC is not merely due to distance reduction by WM in itself. We then provide theoretical insights showing that permutations can change the directions of the singular vectors, but not the singular values, of the weight matrices in each layer. This finding shows that permutations found by WM mainly align the directions of singular vectors associated with large singular values across models. This alignment brings the singular vectors with large singular values, which determine the model functionality, closer between pre-merged and post-merged models, so that the post-merged model retains functionality similar to the pre-merged models, making it easy to satisfy LMC. Finally, we analyze the difference between WM and straight-through estimator (STE), a dataset-dependent permutation search method, and show that WM outperforms STE, especially when merging three or more models.
Abstract:This study presents a new framework for vehicle motion planning and control based on the automatic generation of model predictive controllers (MPC) named MPC Builder. In this framework, several components necessary for MPC, such as models, constraints, and cost functions, are prepared in advance. The MPC Builder then online generates various MPCs according to traffic situations in a unified manner. This scheme enabled us to represent various driving tasks with minimal design effort. The proposed framework was implemented considering the continuation/generalized minimum residual (C/GMRES) method optimization solver, which can reduce computational costs. Finally, numerical experiments on multiple driving scenarios were presented.
Abstract:The paper proposes a computationally feasible method for measuring context-sensitive semantic distance between words. The distance is computed by adaptive scaling of a semantic space. In the semantic space, each word in the vocabulary V is represented by a multi-dimensional vector which is obtained from an English dictionary through a principal component analysis. Given a word set C which specifies a context for measuring word distance, each dimension of the semantic space is scaled up or down according to the distribution of C in the semantic space. In the space thus transformed, distance between words in V becomes dependent on the context C. An evaluation through a word prediction task shows that the proposed measurement successfully extracts the context of a text.